Year: 2023
Type: conference-paper
Abstract: Facial affect recognition draws much interest due to its applications in mental health diagnosis, e-learning, and general health monitoring. The commonly used discrete labelling method can be ambiguous due to information loss and the many combinations of emotions. Conversely, the valence-arousal representation provides a more refined definition. We propose an image-based affect monitoring application with categorical and continuous emotion values using video. The discrete emotion is first determined through a majority voting scheme within a sliding frame window, then a regression model is used to predict the valence-arousal intensities. Finally, variations in the predicted emotion labels are displayed in a graph for monitoring purposes. Our approach managed to achieve the best performance in valence prediction and competitive performance in arousal prediction. The F1-score showed acceptable and highly promising results given that it was trained on much fewer samples compared to existing methods.
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4380874130
